CoRe Dataset

该数据集名为CoRe，包含了来自社会科学和行为科学合作文献综述中的（假设、摘要、标签）三联体，重点关注支持或反驳特定假设的证据。该数据集是基于2019年开始的开放源代码合作综述编制而成，包含了同行评审论文、博客文章和报告等多种类型的文章。数据集被划分为训练集（70%）、验证集（15%）和保留集（15%）。规模上，它包含了69个独特的假设和来自602篇文章的638个三联体。该数据集的任务是科学假设的证据支持。

This dataset, named CoRe, comprises (hypothesis, abstract, label) triplets derived from collaborative literature reviews in social and behavioral sciences, with a core focus on evidence supporting or refuting specific hypotheses. It is compiled from open-source collaborative literature reviews initiated in 2019, covering various types of articles including peer-reviewed papers, blog posts, and reports. The dataset is partitioned into three subsets: a training set (70%), a validation set (15%), and a holdout set (15%). In terms of scale, it contains 69 unique hypotheses and 638 triplets sourced from 602 articles. The downstream task for this dataset is evaluating evidence support for scientific hypotheses.